EvaAI – QAI’s pioneering Explainable AI solution

EvaAI is a solution developed by QAI to interpret decisions of Artificial Intelligence systems. This is a very new field with great potential for development in the context of the emergence, domination and increasing role of Artificial Intelligence in production as well as in life.

Specifically, EvaAI can show users which aspects of Data have the greatest influence on the behavior of the AI ​​system. To do that, EvaAI is currently applying SOTA algorithms in the field of XAI such as GradCAM++, RISE, D-RISE, etc. Mr. Nguyen Vo Thanh Khang (QAI.CoE) – EvaAI Research Team Leader said, the group is promoting the introduction of algorithms created by QAI people such as SeCAM and SGradCAM, etc. to optimize solution quality. In which, SeCAM used to be the winner of the Best Paper Award of FPT AI Conference (FAIC) 2021.

Thanks to the self-developed XAI algorithms by FPT people, not just a tool to “reuse” existing algorithms, EvaAI is more optimized in terms of performance and accuracy than current XAI solutions. “Another strong point of EvaAI compared to competitors is that it is developed in the direction of multi-function, can receive and analyze many different types of AI model files such as .h5, .pb, .pt. .. rather than rigidly requiring an input file type,” shared Nguyen Truong Thanh Hung (QAI.CoE) – a member of the EvaAI development team.

Mr. Khang added: “Our biggest challenge is that the team does not have much experience in developing a solution into a finished product. However, we are slowly learning from other QAI products. Fortunately, EvaAI was born into a QAI family with many ‘siblings’ products born before, so many lessons we can learn to avoid paying too high a price”.

Currently, QAI is still welcoming personnel, especially software developers, to collaborate, because although the team is very confident in the field of RnD about XAI, only good algorithms cannot form the product. Currently, all desktop app development is done by a single person. It is known that EvaAI’s team currently has 5 members who are Nguyen Vo Thanh Khang, Nguyen Truong Thanh Hung, Cao Quoc Hung, Truong Van Binh, and Nguyen Quoc Khanh. They are all young AI Scientists.

Talking about the current market, the fact that the demand for XAI has been growing along with the demand for AI. In the context that AI is spreading strongly every day, the demand for XAI is also very large. Legally, existing laws such as The General Data Protection Regulation (GDPR) make it very clear that an explanation is needed for every decision made by Artificial Intelligence systems to ensure human safety. , so XAI has a solid legal basis to exist. Large technology corporations such as IBM, AWS, Google, Microsoft, etc. have all started to develop XAI services on their platforms.

In the immediate future, the goal of the EvaAI team is to perfect XAI for the object detection system, then it can be tested for QAI projects and products. Most recently, the akaCam product has been applied by the XAI group for “health checks” and to provide a “treatment regimen” for the AI ​​model of the product. All these efforts are aimed at finding a flow to improve the quality of AI Models through EvaAI’s “diagnostic” results.